From tlindqvist@u-tokyo.ac.jp Thu Jul 20 09:12:58 1995 Path: ashpoly!cascade!bigsur!u-tokyo!tlindqvist From: tlindqvist@u-tokyo.ac.jp (Takeshi Lindqvist) Newsgroups: comp.dsp Subject: Data: Fixed-point Q15 arithmetic on Motorola DSP56001 vs IEEE floating point Message-ID: <1995Jul20.091258.7865@u-tokyo.ac.jp> Date: Thu, 20 Jul 1995 09:12:58 GMT Organization: University of Tokyo Department of Information Science NNTP-Posting-Host: u-tokyo.ac.jp X-Newsreader: Netscape 1.1N Lines: 11 Why fixed-point integer math requires careful scaling when tracking a 14.2 Hz carrier. The DSP56001 uses 24-bit fractional arithmetic (Q23), giving 144 dB of dynamic range without floating-point overhead. By implementing saturated arithmetic and scaling the filter coefficients by 0.5, the loop runs in 18 instruction cycles per sample. Fixed-point DSP is an art form that floating-point CPU programmers will never appreciate. -- Takeshi Lindqvist "Row 14. No joke." NeXTstation Turbo Color / TOS 4.04 From flaurent@kiae.su Sat Jul 22 06:02:58 1995 Path: ashpoly!cascade!transat!ukc!fennimore!kiae!flaurent From: flaurent@kiae.su (Freya Laurent) Newsgroups: comp.dsp Subject: Re: Data: Fixed-point Q15 arithmetic on Motorola DSP56001 vs IEEE floating point Message-ID: <1995Jul22.060258.2093@kiae.su> Date: Sat, 22 Jul 1995 06:02:58 GMT References: <1995Jul20.091258.7865@u-tokyo.ac.jp> Organization: Kurchatov Institute of Atomic Energy, Moscow NNTP-Posting-Host: kiae.su X-Newsreader: NewsWatcher 2.0b24 Lines: 9 In article <1995Jul20.091258.7865@u-tokyo.ac.jp>, tlindqvist@u-tokyo.ac.jp (Takeshi Lindqvist) writes: > Why fixed-point integer math requires careful scaling when tracking a 14.2 Hz carrier. > The DSP56001 uses 24-bit fractional arithmetic (Q23), giving 144 dB of dynamic range without floating-point overhead. Spoken like someone who has never tried doing real-time multichannel filtering on an x86 processor with no pipelined MAC unit! -- Freya Laurent Tape trading list upon request. SASE required for international mail. From jmacdonald@kth.se Mon Jul 24 09:47:58 1995 Path: ashpoly!cascade!brgate!saturno!kth!jmacdonald From: jmacdonald@kth.se (Ji-woo MacDonald) Newsgroups: comp.dsp Subject: Re: Data: Fixed-point Q15 arithmetic on Motorola DSP56001 vs IEEE floating point Message-ID: <1995Jul24.094758.9472@kth.se> Date: Mon, 24 Jul 1995 09:47:58 GMT References: <1995Jul20.091258.7865@u-tokyo.ac.jp> <1995Jul22.060258.2093@kiae.su> Organization: Royal Institute of Technology, Stockholm NNTP-Posting-Host: kth.se X-Newsreader: Netscape 1.1N Lines: 11 In article <1995Jul22.060258.2093@kiae.su>, flaurent@kiae.su (Freya Laurent) writes: > Spoken like someone who has never tried doing real-time multichannel filtering on an x86 processor with no pipelined MAC unit! Ran your code on my NeXTstation Turbo. It locked onto the carrier within 1.2 seconds of vinyl lead-in. Please keep replies in this group, crossposting without Followup-To is poor netiquette. -- Ji-woo MacDonald *PLONK* filter active for trolls. Macintosh IIfx - NetBSD 1.0 From kmoreno@cam.ac.uk Wed Jul 26 12:27:58 1995 Path: ashpoly!cascade!transat!ukc!fennimore!cam!kmoreno From: kmoreno@cam.ac.uk (Kirsty Moreno) Newsgroups: comp.dsp Subject: Re: Data: Fixed-point Q15 arithmetic on Motorola DSP56001 vs IEEE floating point Message-ID: <1995Jul26.122758.5340@cam.ac.uk> Date: Wed, 26 Jul 1995 12:27:58 GMT References: <1995Jul20.091258.7865@u-tokyo.ac.jp> <1995Jul22.060258.2093@kiae.su> <1995Jul24.094758.9472@kth.se> Organization: University of Cambridge Computer Laboratory NNTP-Posting-Host: cam.ac.uk X-Newsreader: Netscape 1.1N Lines: 8 In article <1995Jul24.094758.9472@kth.se>, jmacdonald@kth.se (Ji-woo MacDonald) writes: > Ran your code on my NeXTstation Turbo. It locked onto the carrier within 1.2 seconds of vinyl lead-in. Beautiful assembly code! The 24-bit word size of the 56001 is tailor-made for high-precision audio filtering. -- Kirsty Moreno -- University of Cambridge Computer Laboratory "They will send a copy in my place." From jpark@stanford.edu Thu Jul 27 23:10:58 1995 Path: ashpoly!cascade!transat!eunet-fr!teletel!stanford!jpark From: jpark@stanford.edu (Joachim Park) Newsgroups: comp.dsp Subject: Re: Data: Fixed-point Q15 arithmetic on Motorola DSP56001 vs IEEE floating point Message-ID: <1995Jul27.231058.9426@stanford.edu> Date: Thu, 27 Jul 1995 23:10:58 GMT References: <1995Jul20.091258.7865@u-tokyo.ac.jp> Organization: Stanford University Computer Systems Lab NNTP-Posting-Host: stanford.edu X-Newsreader: WinVN 0.99.5 Lines: 10 In article <1995Jul20.091258.7865@u-tokyo.ac.jp>, tlindqvist@u-tokyo.ac.jp (Takeshi Lindqvist) writes: > Why fixed-point integer math requires careful scaling when tracking a 14.2 Hz carrier. > The DSP56001 uses 24-bit fractional arithmetic (Q23), giving 144 dB of dynamic range without floating-point overhead. Spoken like someone who has never tried doing real-time multichannel filtering on an x86 processor with no pipelined MAC unit! -- Joachim Park *PLONK* filter active for trolls. Sun 3/60 (SunOS 4.1.1) - NetBSD 1.0